• Home
• Education
• Projects
• Miscellaneous
• Contact

Contents

Brain tumor detector

Nutshell

In this project I build a program that detects and localizes cancer from images of human brains, as explained on the course Modern Artificial Intelligence, lectured by Dr. Ryan Ahmed, Ph.D. MBA.

I will train two models which will

1. classify the images either containing cancer tumor or not
2. localizes the tumor within the brain

Introduction to the Brain Tumor Detection

Deep learning has proven to be as good and even better than humans in detecting diseases from X-rays, MRI scans and CT scans. there is huge potential in using AI to speed up and improve the accuracy of diagnosis. This project will use the labeled dataset from https://www.kaggle.com/datasets/mateuszbuda/lgg-mri-segmentation which consists of 3929 Brain MRI scans and the tumor location. The final pipeline has a two step process where

1. A Resnet deep learning classifier model will classify the input images into two groups: tumor detected and tumor not detected.
2. For the images, where tumor was detected, a second step is performed, where a ResUNet segmentation model detects the tumor location on the pixel level.

Image segmentation

Image segmentation extracts information from images on the level of pixels. It is used for object recognition and localization in applications like medical imaging and self-driving cars. Image segmentation produces a pixel-wise mask of the image with deep learning approaches using common architectures such as CNN, FNNs and Deep Encoders-Decoders.

With Unet, the input and the output have the same size so the size of the images is preserved. In contrast to the CNN image classification, where the image is converted to a vector and the entire image is classified as a class label, the Unet performs classification on pixel level. Unet formulates a loss function for every pixel and then a softmax function is applied to every pixel. In other words, the segmentation problem is solved as a classification problem.

Looking into the data

We have a csv file that contains the patient IDs, the locations of the images, their masks and indicator if there is a tumor in the image (1 - tumor, 0 - healthy). There are 1373 images with tumors and 2556 healthy brain images. Thus, the dataset is imbalanced.

<class 'pandas.core.frame.DataFrame'>
RangeIndex: 3929 entries, 0 to 3928
Data columns (total 4 columns):
 #   Column      Non-Null Count  Dtype 
---  ------      --------------  ----- 
 0   patient_id  3929 non-null   object
 1   image_path  3929 non-null   object
 2   mask_path   3929 non-null   object
 3   mask        3929 non-null   int64 
dtypes: int64(1), object(3)
memory usage: 122.9+ KB

	patient_id	image_path	mask_path	mask
0	TCGA_CS_5395_19981004	TCGA_CS_5395_19981004/TCGA_CS_5395_19981004_1.tif	TCGA_CS_5395_19981004/TCGA_CS_5395_19981004_1_...	0
1	TCGA_CS_5395_19981004	TCGA_CS_4944_20010208/TCGA_CS_4944_20010208_1.tif	TCGA_CS_4944_20010208/TCGA_CS_4944_20010208_1_...	0
2	TCGA_CS_5395_19981004	TCGA_CS_4941_19960909/TCGA_CS_4941_19960909_1.tif	TCGA_CS_4941_19960909/TCGA_CS_4941_19960909_1_...	0
3	TCGA_CS_5395_19981004	TCGA_CS_4943_20000902/TCGA_CS_4943_20000902_1.tif	TCGA_CS_4943_20000902/TCGA_CS_4943_20000902_1_...	0
4	TCGA_CS_5395_19981004	TCGA_CS_5396_20010302/TCGA_CS_5396_20010302_1.tif	TCGA_CS_5396_20010302/TCGA_CS_5396_20010302_1_...	0

Visualisation of the datasets

Below is an example of an MRI image and the matching mask. This example has a small tumor. In images where no tumor is present, the mask will be completely black.

Below are visualisations from 6 MRIs and their overlayed masks in rose color to get a sense of the data that I will be using in this project.

Convolutional neural networks (CNNs)

• The first CNN layers are used to extract high level general features
• The last couple of layers will perform classification
• Locla respective fields scan the image first searching for simple shapes such as edges and lines
• The edges are picked up by the subsequent layer to form more complex features

A good visualisation of the feature extraction with convolutions can be found at https://setosa.io/ev/image-kernels/

ResNet (Residual Network)

• As CNNs grow deeper, vanishing gradients negatively impact the network performance. Vanishing gradient occurs when the gradient is backpropagated to earlier layers which results in a very small gradient.
• ResNets "skip connection" feature can allow training of 152 layers without vanishing gradient problems
• ResNet adds "identity mapping on top of the CNN
• ResNet deep network is trained with ImageNet, which contains 11 million images and 11 000 categories

ResNet paper (He etal, 2015): https://arxiv.org/pdf/1512.03385

As seen in the Figure 6. from the Resnet paper, the ResNet architectures overcome the training challenges from deep networks compared ot the plain networks. ResNet-152 achieved 3.58% error rate on the ImageNet dataset. This is better than human performance.

Siddarth Das has made a great comparison of CNN architecture performances, you can check it out here: https://medium.com/analytics-vidhya/cnns-architectures-lenet-alexnet-vgg-googlenet-resnet-and-more-666091488df5

Transfer learning

Transfer learning retrains a network that has been trained to perform a specific task to use it in a similar task. The use of a pretrained model can drastically reduce the computational time and the amount of training data required, compared to starting from scratch. It can be compared to a salsa dancer starting to learn bachata; he/she will probably do a lot better than a person who has never danced before.

There are two main strategies in transfer learning:

1. Freeze the trained CNN network weights from the first layers and the train newly added dense layers. The new layers are initialized with random weights.
2. Retrain the entire CNN network while setting the learning rate to be very small. With too large learning rate the already trained weights might be changed too dramatically.

In this project I will use the approach 1.

Transfer learning has it's own challenges:

• Negative Transfer: the source task/domain is “close enough to look useful” but actually pushes the model in the wrong direction, hurting performance compared to training from scratch. This occurs when the features of old and new tasks are not related.
• Which layers to transfer / freeze: deciding what to reuse vs retrain is nontrivial; freezing too much can underfit, unfreezing too much can overfit or destabilize training.
• Representation misalignment: even if tasks are related, the internal features might not separate target classes well, especially when target cues differ (e.g., medical imaging vs natural images).
• Transfer bounds: Measuring the amount of knowledge transferred is crucial to ensure model quality and robustness. It is worth considering, how to quantify this, and it is a subject of ongoing research.

This is a great resource for transfer learning from Dipanjan Sarkar: https://towardsdatascience.com/a-comprehensive-hands-on-guide-to-transfer-learning-with-real-world-applications-in-deep-learning-212bf3b2f27a/

ResUNet

I will use ResUNet in the second part for the segmentation of the tumors.

• ResUNet architecture combines UNet backbone architecture and residual blocks
• The Unet uses Fully Convolutional Networks (FCN) and is adapted to perform well on segmentation tasks
• ResUNet has three parts:
  1. Encoder or contracting path
  2. Bottleneck
  3. Decoder or expansive path

The contraction path consists of several contraction blocks, which pass their input through res-blocks followed by 2x2 max-pooling. Feature maps after each block doubles, which helps the model learn complex features effectively.

The bottleneck part takes the input and then passes through a resblock, followed by 2x2 up-sampling convolution layers.

The decoder blocks take the up-sampled input from the previous layer and concatenates with the corresponding output features from the res-blocks in the contraction path. This is then passed through a resblock. This ensures that the features learned while contracting are used while reconstructing the image.

The final expansion layer output is passed through 1x1 convolution layer to produce the desired output with the same size as the input.

The original paper that introduced ResUNet: https://arxiv.org/pdf/1904.00592

Part 1: Training a classifier model to detect if tumor exists or not

I use the flow_from_dataframe for training. Batch size = 16 class mode = categorical

# @title Preparing image generators
train_generator = datagen.flow_from_dataframe(
    dataframe = train,
    directory = './',
    x_col = 'image_path',
    y_col = 'mask',
    subset = 'training',
    batch_size =16,
    shuffle = True,
    class_mode = 'categorical',
    target_size = (256, 256)
)
valid_generator = datagen.flow_from_dataframe(
    dataframe = train,
    directory = './',
    x_col = 'image_path',
    y_col = 'mask',
    subset = 'validation',
    batch_size = 16,
    shuffle = True,
    class_mode = 'categorical',
    target_size = (256, 256)
)
#create a data generator for test images
#no need for splitting again because here we use the "test" data set
test_datagen = ImageDataGenerator(rescale=1./255)

test_generator = datagen.flow_from_dataframe(
    dataframe = test,
    directory = './',
    x_col = 'image_path',
    y_col = 'mask',
    batch_size = 16,
    shuffle = False,
    class_mode = 'categorical',
    target_size = (256, 256)
)

Found 2839 validated image filenames belonging to 2 classes.
Found 500 validated image filenames belonging to 2 classes.
Found 590 validated image filenames belonging to 2 classes.

Below is the architecture of the ResNet50 model. For the transfer learning, all of these layers will be set to trainable = False to stop the weights from changing. The last layers in purple are the added layers which will be trained.

# @title Retrieve ResNet50 base model
#Input tensror 256 x 256 x 3
basemodel = ResNet50(weights = 'imagenet', include_top = False,
                     input_tensor = Input(shape = (256, 256, 3)))

Downloading data from https://storage.googleapis.com/tensorflow/keras-applications/resnet/resnet50_weights_tf_dim_ordering_tf_kernels_notop.h5
94765736/94765736 ━━━━━━━━━━━━━━━━━━━━ 0s 0us/step

# Add classification head to the base model
headmodel = basemodel.output
headmodel = AveragePooling2D(pool_size = (4, 4))(headmodel)
headmodel = Flatten(name = 'flatten')(headmodel)
headmodel = Dense(256, activation = 'relu')(headmodel)
headmodel = Dropout(0.3)(headmodel)
headmodel = Dense(2, activation = 'softmax')(headmodel)

fullmodel = Model(inputs = basemodel.input, outputs = headmodel)

# compile the model
fullmodel.compile(loss = 'categorical_crossentropy', optimizer='adam',
              metrics=["accuracy"])

# use the early stopping to exit training
earlystopping = EarlyStopping(monitor='val_loss',
                              mode='min',
                              verbose = 1,
                              patience = 20)
# save the best model with least validation loss
checkpointer = ModelCheckpoint(filepath=model_base/'classifier-resnet-weights_22-12-2025.keras',
                               verbose=1,
                               save_best_only=True)

# Callbacks: logs epoch results to CSV
csv_logger = CSVLogger(
    model_base/"training_history_classifier_model_22-12-2025.csv",
    append=True,         # keep adding if file exists
    separator=','        # comma-separated
    )

if train_model:
  history = fullmodel.fit(train_generator,
              steps_per_epoch = train_generator.n // train_generator.batch_size,
              epochs=60,
              validation_data=valid_generator,
              validation_steps= valid_generator.n // valid_generator.batch_size,
              callbacks=[checkpointer, earlystopping,csv_logger])

/usr/local/lib/python3.12/dist-packages/keras/src/trainers/data_adapters/py_dataset_adapter.py:121: UserWarning:

Your `PyDataset` class should call `super().__init__(**kwargs)` in its constructor. `**kwargs` can include `workers`, `use_multiprocessing`, `max_queue_size`. Do not pass these arguments to `fit()`, as they will be ignored.

Epoch 1/60
177/177 ━━━━━━━━━━━━━━━━━━━━ 0s 5s/step - accuracy: 0.6593 - loss: 1.2657
Epoch 1: val_loss improved from inf to 1.52948, saving model to /content/drive/MyDrive/Colab Notebooks/brain-tumor-detector/Models/classifier-resnet-weights_22-12-2025.keras
177/177 ━━━━━━━━━━━━━━━━━━━━ 1071s 6s/step - accuracy: 0.6596 - loss: 1.2629 - val_accuracy: 0.3367 - val_loss: 1.5295
Epoch 2/60
  1/177 ━━━━━━━━━━━━━━━━━━━━ 8s 48ms/step - accuracy: 0.8750 - loss: 0.2890

/usr/local/lib/python3.12/dist-packages/keras/src/trainers/epoch_iterator.py:116: UserWarning:

Your input ran out of data; interrupting training. Make sure that your dataset or generator can generate at least `steps_per_epoch * epochs` batches. You may need to use the `.repeat()` function when building your dataset.

Epoch 2: val_loss did not improve from 1.52948
177/177 ━━━━━━━━━━━━━━━━━━━━ 2s 10ms/step - accuracy: 0.8750 - loss: 0.2890 - val_accuracy: 0.3387 - val_loss: 1.8686
Epoch 3/60
177/177 ━━━━━━━━━━━━━━━━━━━━ 0s 57ms/step - accuracy: 0.7892 - loss: 0.4580
Epoch 3: val_loss improved from 1.52948 to 0.65035, saving model to /content/drive/MyDrive/Colab Notebooks/brain-tumor-detector/Models/classifier-resnet-weights_22-12-2025.keras
177/177 ━━━━━━━━━━━━━━━━━━━━ 14s 77ms/step - accuracy: 0.7893 - loss: 0.4579 - val_accuracy: 0.6593 - val_loss: 0.6504
Epoch 4/60
  1/177 ━━━━━━━━━━━━━━━━━━━━ 7s 45ms/step - accuracy: 0.9375 - loss: 0.1734
Epoch 4: val_loss improved from 0.65035 to 0.64082, saving model to /content/drive/MyDrive/Colab Notebooks/brain-tumor-detector/Models/classifier-resnet-weights_22-12-2025.keras
177/177 ━━━━━━━━━━━━━━━━━━━━ 4s 24ms/step - accuracy: 0.9375 - loss: 0.1734 - val_accuracy: 0.6613 - val_loss: 0.6408
Epoch 5/60
177/177 ━━━━━━━━━━━━━━━━━━━━ 0s 93ms/step - accuracy: 0.8342 - loss: 0.4006
Epoch 5: val_loss did not improve from 0.64082
177/177 ━━━━━━━━━━━━━━━━━━━━ 18s 103ms/step - accuracy: 0.8342 - loss: 0.4006 - val_accuracy: 0.6633 - val_loss: 2.8768
Epoch 6/60
  1/177 ━━━━━━━━━━━━━━━━━━━━ 7s 43ms/step - accuracy: 0.9375 - loss: 0.2511
Epoch 6: val_loss did not improve from 0.64082
177/177 ━━━━━━━━━━━━━━━━━━━━ 2s 10ms/step - accuracy: 0.9375 - loss: 0.2511 - val_accuracy: 0.6593 - val_loss: 3.2475
Epoch 7/60
177/177 ━━━━━━━━━━━━━━━━━━━━ 0s 56ms/step - accuracy: 0.8630 - loss: 0.3163
Epoch 7: val_loss did not improve from 0.64082
177/177 ━━━━━━━━━━━━━━━━━━━━ 12s 66ms/step - accuracy: 0.8631 - loss: 0.3162 - val_accuracy: 0.6593 - val_loss: 1.0518
Epoch 8/60
  1/177 ━━━━━━━━━━━━━━━━━━━━ 7s 42ms/step - accuracy: 0.7500 - loss: 0.3713
Epoch 8: val_loss did not improve from 0.64082
177/177 ━━━━━━━━━━━━━━━━━━━━ 2s 9ms/step - accuracy: 0.7500 - loss: 0.3713 - val_accuracy: 0.6633 - val_loss: 1.0338
Epoch 9/60
177/177 ━━━━━━━━━━━━━━━━━━━━ 0s 55ms/step - accuracy: 0.8790 - loss: 0.2813
Epoch 9: val_loss improved from 0.64082 to 0.61995, saving model to /content/drive/MyDrive/Colab Notebooks/brain-tumor-detector/Models/classifier-resnet-weights_22-12-2025.keras
177/177 ━━━━━━━━━━━━━━━━━━━━ 13s 75ms/step - accuracy: 0.8791 - loss: 0.2813 - val_accuracy: 0.6935 - val_loss: 0.6200
Epoch 10/60
  1/177 ━━━━━━━━━━━━━━━━━━━━ 7s 44ms/step - accuracy: 0.8750 - loss: 0.4855
Epoch 10: val_loss improved from 0.61995 to 0.60929, saving model to /content/drive/MyDrive/Colab Notebooks/brain-tumor-detector/Models/classifier-resnet-weights_22-12-2025.keras
177/177 ━━━━━━━━━━━━━━━━━━━━ 5s 25ms/step - accuracy: 0.8750 - loss: 0.4855 - val_accuracy: 0.7036 - val_loss: 0.6093
Epoch 11/60
177/177 ━━━━━━━━━━━━━━━━━━━━ 0s 88ms/step - accuracy: 0.9103 - loss: 0.2541
Epoch 11: val_loss improved from 0.60929 to 0.43979, saving model to /content/drive/MyDrive/Colab Notebooks/brain-tumor-detector/Models/classifier-resnet-weights_22-12-2025.keras
177/177 ━━━━━━━━━━━━━━━━━━━━ 20s 108ms/step - accuracy: 0.9103 - loss: 0.2541 - val_accuracy: 0.7964 - val_loss: 0.4398
Epoch 12/60
  1/177 ━━━━━━━━━━━━━━━━━━━━ 7s 45ms/step - accuracy: 1.0000 - loss: 0.0712
Epoch 12: val_loss improved from 0.43979 to 0.42589, saving model to /content/drive/MyDrive/Colab Notebooks/brain-tumor-detector/Models/classifier-resnet-weights_22-12-2025.keras
177/177 ━━━━━━━━━━━━━━━━━━━━ 4s 20ms/step - accuracy: 1.0000 - loss: 0.0712 - val_accuracy: 0.8145 - val_loss: 0.4259
Epoch 13/60
177/177 ━━━━━━━━━━━━━━━━━━━━ 0s 94ms/step - accuracy: 0.9239 - loss: 0.2116
Epoch 13: val_loss improved from 0.42589 to 0.27524, saving model to /content/drive/MyDrive/Colab Notebooks/brain-tumor-detector/Models/classifier-resnet-weights_22-12-2025.keras
177/177 ━━━━━━━━━━━━━━━━━━━━ 20s 114ms/step - accuracy: 0.9239 - loss: 0.2115 - val_accuracy: 0.8911 - val_loss: 0.2752
Epoch 14/60
  1/177 ━━━━━━━━━━━━━━━━━━━━ 7s 42ms/step - accuracy: 0.9375 - loss: 0.1121
Epoch 14: val_loss did not improve from 0.27524
177/177 ━━━━━━━━━━━━━━━━━━━━ 2s 10ms/step - accuracy: 0.9375 - loss: 0.1121 - val_accuracy: 0.8851 - val_loss: 0.2856
Epoch 15/60
177/177 ━━━━━━━━━━━━━━━━━━━━ 0s 88ms/step - accuracy: 0.9442 - loss: 0.1573
Epoch 15: val_loss improved from 0.27524 to 0.26114, saving model to /content/drive/MyDrive/Colab Notebooks/brain-tumor-detector/Models/classifier-resnet-weights_22-12-2025.keras
177/177 ━━━━━━━━━━━━━━━━━━━━ 19s 107ms/step - accuracy: 0.9442 - loss: 0.1573 - val_accuracy: 0.9032 - val_loss: 0.2611
Epoch 16/60
  1/177 ━━━━━━━━━━━━━━━━━━━━ 7s 43ms/step - accuracy: 0.9375 - loss: 0.1289
Epoch 16: val_loss improved from 0.26114 to 0.26042, saving model to /content/drive/MyDrive/Colab Notebooks/brain-tumor-detector/Models/classifier-resnet-weights_22-12-2025.keras
177/177 ━━━━━━━━━━━━━━━━━━━━ 4s 24ms/step - accuracy: 0.9375 - loss: 0.1289 - val_accuracy: 0.9073 - val_loss: 0.2604
Epoch 17/60
177/177 ━━━━━━━━━━━━━━━━━━━━ 0s 92ms/step - accuracy: 0.9408 - loss: 0.1535
Epoch 17: val_loss improved from 0.26042 to 0.18035, saving model to /content/drive/MyDrive/Colab Notebooks/brain-tumor-detector/Models/classifier-resnet-weights_22-12-2025.keras
177/177 ━━━━━━━━━━━━━━━━━━━━ 20s 112ms/step - accuracy: 0.9409 - loss: 0.1534 - val_accuracy: 0.9415 - val_loss: 0.1803
Epoch 18/60
  1/177 ━━━━━━━━━━━━━━━━━━━━ 7s 42ms/step - accuracy: 1.0000 - loss: 0.0220
Epoch 18: val_loss did not improve from 0.18035
177/177 ━━━━━━━━━━━━━━━━━━━━ 2s 11ms/step - accuracy: 1.0000 - loss: 0.0220 - val_accuracy: 0.9395 - val_loss: 0.1892
Epoch 19/60
177/177 ━━━━━━━━━━━━━━━━━━━━ 0s 89ms/step - accuracy: 0.9618 - loss: 0.1116
Epoch 19: val_loss improved from 0.18035 to 0.14857, saving model to /content/drive/MyDrive/Colab Notebooks/brain-tumor-detector/Models/classifier-resnet-weights_22-12-2025.keras
177/177 ━━━━━━━━━━━━━━━━━━━━ 19s 109ms/step - accuracy: 0.9618 - loss: 0.1116 - val_accuracy: 0.9395 - val_loss: 0.1486
Epoch 20/60
  1/177 ━━━━━━━━━━━━━━━━━━━━ 7s 43ms/step - accuracy: 0.9375 - loss: 0.0897
Epoch 20: val_loss improved from 0.14857 to 0.14670, saving model to /content/drive/MyDrive/Colab Notebooks/brain-tumor-detector/Models/classifier-resnet-weights_22-12-2025.keras
177/177 ━━━━━━━━━━━━━━━━━━━━ 3s 19ms/step - accuracy: 0.9375 - loss: 0.0897 - val_accuracy: 0.9395 - val_loss: 0.1467
Epoch 21/60
177/177 ━━━━━━━━━━━━━━━━━━━━ 0s 93ms/step - accuracy: 0.9781 - loss: 0.0578
Epoch 21: val_loss did not improve from 0.14670
177/177 ━━━━━━━━━━━━━━━━━━━━ 18s 103ms/step - accuracy: 0.9780 - loss: 0.0580 - val_accuracy: 0.8105 - val_loss: 0.4949
Epoch 22/60
  1/177 ━━━━━━━━━━━━━━━━━━━━ 7s 43ms/step - accuracy: 0.9375 - loss: 0.0745
Epoch 22: val_loss did not improve from 0.14670
177/177 ━━━━━━━━━━━━━━━━━━━━ 2s 10ms/step - accuracy: 0.9375 - loss: 0.0745 - val_accuracy: 0.8145 - val_loss: 0.5055
Epoch 23/60
177/177 ━━━━━━━━━━━━━━━━━━━━ 0s 69ms/step - accuracy: 0.9600 - loss: 0.1070
Epoch 23: val_loss did not improve from 0.14670
177/177 ━━━━━━━━━━━━━━━━━━━━ 14s 79ms/step - accuracy: 0.9599 - loss: 0.1072 - val_accuracy: 0.6552 - val_loss: 29.7710
Epoch 24/60
  1/177 ━━━━━━━━━━━━━━━━━━━━ 7s 43ms/step - accuracy: 1.0000 - loss: 0.1288
Epoch 24: val_loss did not improve from 0.14670
177/177 ━━━━━━━━━━━━━━━━━━━━ 2s 9ms/step - accuracy: 1.0000 - loss: 0.1288 - val_accuracy: 0.6573 - val_loss: 30.7215
Epoch 25/60
177/177 ━━━━━━━━━━━━━━━━━━━━ 0s 56ms/step - accuracy: 0.9227 - loss: 0.2066
Epoch 25: val_loss did not improve from 0.14670
177/177 ━━━━━━━━━━━━━━━━━━━━ 12s 65ms/step - accuracy: 0.9227 - loss: 0.2066 - val_accuracy: 0.8790 - val_loss: 1.6968
Epoch 26/60
  1/177 ━━━━━━━━━━━━━━━━━━━━ 7s 43ms/step - accuracy: 0.9375 - loss: 0.4549
Epoch 26: val_loss did not improve from 0.14670
177/177 ━━━━━━━━━━━━━━━━━━━━ 2s 9ms/step - accuracy: 0.9375 - loss: 0.4549 - val_accuracy: 0.8589 - val_loss: 2.0143
Epoch 27/60
177/177 ━━━━━━━━━━━━━━━━━━━━ 0s 55ms/step - accuracy: 0.9463 - loss: 0.1448
Epoch 27: val_loss did not improve from 0.14670
177/177 ━━━━━━━━━━━━━━━━━━━━ 11s 64ms/step - accuracy: 0.9463 - loss: 0.1448 - val_accuracy: 0.9274 - val_loss: 0.2727
Epoch 28/60
  1/177 ━━━━━━━━━━━━━━━━━━━━ 6s 38ms/step - accuracy: 0.9375 - loss: 0.1148
Epoch 28: val_loss did not improve from 0.14670
177/177 ━━━━━━━━━━━━━━━━━━━━ 2s 10ms/step - accuracy: 0.9375 - loss: 0.1148 - val_accuracy: 0.9234 - val_loss: 0.2679
Epoch 29/60
177/177 ━━━━━━━━━━━━━━━━━━━━ 0s 55ms/step - accuracy: 0.9645 - loss: 0.0904
Epoch 29: val_loss did not improve from 0.14670
177/177 ━━━━━━━━━━━━━━━━━━━━ 11s 64ms/step - accuracy: 0.9645 - loss: 0.0906 - val_accuracy: 0.7742 - val_loss: 1.2347
Epoch 30/60
  1/177 ━━━━━━━━━━━━━━━━━━━━ 6s 38ms/step - accuracy: 0.9375 - loss: 0.1273
Epoch 30: val_loss did not improve from 0.14670
177/177 ━━━━━━━━━━━━━━━━━━━━ 2s 9ms/step - accuracy: 0.9375 - loss: 0.1273 - val_accuracy: 0.7903 - val_loss: 1.2631
Epoch 31/60
177/177 ━━━━━━━━━━━━━━━━━━━━ 0s 58ms/step - accuracy: 0.9619 - loss: 0.1006
Epoch 31: val_loss did not improve from 0.14670
177/177 ━━━━━━━━━━━━━━━━━━━━ 12s 68ms/step - accuracy: 0.9619 - loss: 0.1007 - val_accuracy: 0.9214 - val_loss: 0.2919
Epoch 32/60
  1/177 ━━━━━━━━━━━━━━━━━━━━ 7s 43ms/step - accuracy: 0.9375 - loss: 0.3318
Epoch 32: val_loss did not improve from 0.14670
177/177 ━━━━━━━━━━━━━━━━━━━━ 2s 10ms/step - accuracy: 0.9375 - loss: 0.3318 - val_accuracy: 0.9153 - val_loss: 0.2921
Epoch 33/60
177/177 ━━━━━━━━━━━━━━━━━━━━ 0s 56ms/step - accuracy: 0.9603 - loss: 0.1285
Epoch 33: val_loss did not improve from 0.14670
177/177 ━━━━━━━━━━━━━━━━━━━━ 12s 65ms/step - accuracy: 0.9602 - loss: 0.1288 - val_accuracy: 0.8911 - val_loss: 0.4200
Epoch 34/60
  1/177 ━━━━━━━━━━━━━━━━━━━━ 6s 38ms/step - accuracy: 0.9375 - loss: 0.3122
Epoch 34: val_loss did not improve from 0.14670
177/177 ━━━━━━━━━━━━━━━━━━━━ 2s 10ms/step - accuracy: 0.9375 - loss: 0.3122 - val_accuracy: 0.8911 - val_loss: 0.4037
Epoch 35/60
177/177 ━━━━━━━━━━━━━━━━━━━━ 0s 55ms/step - accuracy: 0.9632 - loss: 0.0953
Epoch 35: val_loss did not improve from 0.14670
177/177 ━━━━━━━━━━━━━━━━━━━━ 12s 65ms/step - accuracy: 0.9632 - loss: 0.0953 - val_accuracy: 0.9415 - val_loss: 0.1992
Epoch 36/60
  1/177 ━━━━━━━━━━━━━━━━━━━━ 6s 39ms/step - accuracy: 1.0000 - loss: 0.0069
Epoch 36: val_loss did not improve from 0.14670
177/177 ━━━━━━━━━━━━━━━━━━━━ 2s 10ms/step - accuracy: 1.0000 - loss: 0.0069 - val_accuracy: 0.9456 - val_loss: 0.1761
Epoch 37/60
177/177 ━━━━━━━━━━━━━━━━━━━━ 0s 55ms/step - accuracy: 0.9762 - loss: 0.0785
Epoch 37: val_loss did not improve from 0.14670
177/177 ━━━━━━━━━━━━━━━━━━━━ 11s 65ms/step - accuracy: 0.9762 - loss: 0.0785 - val_accuracy: 0.9355 - val_loss: 0.2020
Epoch 38/60
  1/177 ━━━━━━━━━━━━━━━━━━━━ 6s 38ms/step - accuracy: 1.0000 - loss: 0.0051
Epoch 38: val_loss did not improve from 0.14670
177/177 ━━━━━━━━━━━━━━━━━━━━ 2s 10ms/step - accuracy: 1.0000 - loss: 0.0051 - val_accuracy: 0.9355 - val_loss: 0.1975
Epoch 39/60
177/177 ━━━━━━━━━━━━━━━━━━━━ 0s 54ms/step - accuracy: 0.9759 - loss: 0.0767
Epoch 39: val_loss did not improve from 0.14670
177/177 ━━━━━━━━━━━━━━━━━━━━ 11s 64ms/step - accuracy: 0.9759 - loss: 0.0769 - val_accuracy: 0.8891 - val_loss: 0.2712
Epoch 40/60
  1/177 ━━━━━━━━━━━━━━━━━━━━ 7s 43ms/step - accuracy: 1.0000 - loss: 0.0333
Epoch 40: val_loss did not improve from 0.14670
177/177 ━━━━━━━━━━━━━━━━━━━━ 2s 10ms/step - accuracy: 1.0000 - loss: 0.0333 - val_accuracy: 0.8931 - val_loss: 0.2635
Epoch 40: early stopping

The best model was reached with 20 epochs, after which the training was terminated by early stopping (patience=20). Interestingly, the validation loss had one massive peak.

Assess classifier model performance

/usr/local/lib/python3.12/dist-packages/keras/src/trainers/data_adapters/py_dataset_adapter.py:121: UserWarning:

Your `PyDataset` class should call `super().__init__(**kwargs)` in its constructor. `**kwargs` can include `workers`, `use_multiprocessing`, `max_queue_size`. Do not pass these arguments to `fit()`, as they will be ignored.

37/37 ━━━━━━━━━━━━━━━━━━━━ 184s 5s/step

The model accuracy is 0.94

from sklearn.metrics import classification_report
print(classification_report(y_true, y_pred, labels = [0,1]))

              precision    recall  f1-score   support

           0       0.93      0.98      0.95       366
           1       0.97      0.88      0.92       224

   micro avg       0.94      0.94      0.94       590
   macro avg       0.95      0.93      0.94       590
weighted avg       0.94      0.94      0.94       590

Part 2: Building a segmentation model to localise tumors

(1373, 4)

Saved to: /content/drive/MyDrive/Colab Notebooks/brain-tumor-detector/Models/segmentation_splits.json

def resblock(X, f):


  # make a copy of input
  X_copy = X

  # main path
  # Read more about he_normal: https://medium.com/@prateekvishnu/xavier-and-he-normal-he-et-al-initialization-8e3d7a087528

  X = Conv2D(f, kernel_size = (1,1) ,strides = (1,1),kernel_initializer ='he_normal')(X)
  X = BatchNormalization()(X)
  X = Activation('relu')(X)

  X = Conv2D(f, kernel_size = (3,3), strides =(1,1), padding = 'same', kernel_initializer ='he_normal')(X)
  X = BatchNormalization()(X)

  # Short path
  # Read more here: https://towardsdatascience.com/understanding-and-coding-a-resnet-in-keras-446d7ff84d33

  X_copy = Conv2D(f, kernel_size = (1,1), strides =(1,1), kernel_initializer ='he_normal')(X_copy)
  X_copy = BatchNormalization()(X_copy)

  # Adding the output from main path and short path together

  X = Add()([X,X_copy])
  X = Activation('relu')(X)

  return X

# function to upscale and concatenate the values passed
def upsample_concat(x, skip):
  x = UpSampling2D((2,2))(x)
  merge = Concatenate()([x, skip])

  return merge

input_shape = (256,256,3)

# Input tensor shape
X_input = Input(input_shape)

# Stage 1
conv1_in = Conv2D(16,3,activation= 'relu', padding = 'same', kernel_initializer ='he_normal')(X_input)
conv1_in = BatchNormalization()(conv1_in)
conv1_in = Conv2D(16,3,activation= 'relu', padding = 'same', kernel_initializer ='he_normal')(conv1_in)
conv1_in = BatchNormalization()(conv1_in)
pool_1 = MaxPool2D(pool_size = (2,2))(conv1_in)

# Stage 2
conv2_in = resblock(pool_1, 32)
pool_2 = MaxPool2D(pool_size = (2,2))(conv2_in)

# Stage 3
conv3_in = resblock(pool_2, 64)
pool_3 = MaxPool2D(pool_size = (2,2))(conv3_in)

# Stage 4
conv4_in = resblock(pool_3, 128)
pool_4 = MaxPool2D(pool_size = (2,2))(conv4_in)

# Stage 5 (Bottle Neck)
conv5_in = resblock(pool_4, 256)

# Upscale stage 1
up_1 = upsample_concat(conv5_in, conv4_in)
up_1 = resblock(up_1, 128)

# Upscale stage 2
up_2 = upsample_concat(up_1, conv3_in)
up_2 = resblock(up_2, 64)

# Upscale stage 3
up_3 = upsample_concat(up_2, conv2_in)
up_3 = resblock(up_3, 32)

# Upscale stage 4
up_4 = upsample_concat(up_3, conv1_in)
up_4 = resblock(up_4, 16)

# Final Output
output = Conv2D(1, (1,1), padding = "same", activation = "sigmoid")(up_4)

model_seg = Model(inputs = X_input, outputs = output )

# @title Compiling the segmentation model
from utilities import tversky, tversky_loss, focal_tversky
adam = keras.optimizers.Adam(learning_rate = 0.05, epsilon = 0.1)
model_seg.compile(optimizer = adam, loss = focal_tversky, metrics = [tversky])

/usr/local/lib/python3.12/dist-packages/keras/src/trainers/data_adapters/py_dataset_adapter.py:121: UserWarning:

Your `PyDataset` class should call `super().__init__(**kwargs)` in its constructor. `**kwargs` can include `workers`, `use_multiprocessing`, `max_queue_size`. Do not pass these arguments to `fit()`, as they will be ignored.

Epoch 1/60
72/72 ━━━━━━━━━━━━━━━━━━━━ 0s 9s/step - loss: 0.8267 - tversky: 0.2203

/usr/local/lib/python3.12/dist-packages/keras/src/trainers/data_adapters/py_dataset_adapter.py:121: UserWarning:

Your `PyDataset` class should call `super().__init__(**kwargs)` in its constructor. `**kwargs` can include `workers`, `use_multiprocessing`, `max_queue_size`. Do not pass these arguments to `fit()`, as they will be ignored.

Epoch 1: val_loss improved from inf to 0.59689, saving model to /content/drive/MyDrive/Colab Notebooks/brain-tumor-detector/Models/ResUNet-weights_22-12-2025.keras
72/72 ━━━━━━━━━━━━━━━━━━━━ 760s 10s/step - loss: 0.8249 - tversky: 0.2224 - val_loss: 0.5969 - val_tversky: 0.4970
Epoch 2/60
72/72 ━━━━━━━━━━━━━━━━━━━━ 0s 259ms/step - loss: 0.4331 - tversky: 0.6690
Epoch 2: val_loss did not improve from 0.59689
72/72 ━━━━━━━━━━━━━━━━━━━━ 22s 304ms/step - loss: 0.4330 - tversky: 0.6691 - val_loss: 0.6728 - val_tversky: 0.4079
Epoch 3/60
72/72 ━━━━━━━━━━━━━━━━━━━━ 0s 144ms/step - loss: 0.3882 - tversky: 0.7145
Epoch 3: val_loss improved from 0.59689 to 0.36208, saving model to /content/drive/MyDrive/Colab Notebooks/brain-tumor-detector/Models/ResUNet-weights_22-12-2025.keras
72/72 ━━━━━━━━━━━━━━━━━━━━ 12s 168ms/step - loss: 0.3879 - tversky: 0.7147 - val_loss: 0.3621 - val_tversky: 0.7410
Epoch 4/60
72/72 ━━━━━━━━━━━━━━━━━━━━ 0s 148ms/step - loss: 0.3287 - tversky: 0.7716
Epoch 4: val_loss did not improve from 0.36208
72/72 ━━━━━━━━━━━━━━━━━━━━ 12s 161ms/step - loss: 0.3287 - tversky: 0.7716 - val_loss: 0.3912 - val_tversky: 0.7129
Epoch 5/60
72/72 ━━━━━━━━━━━━━━━━━━━━ 0s 147ms/step - loss: 0.2861 - tversky: 0.8098
Epoch 5: val_loss improved from 0.36208 to 0.34319, saving model to /content/drive/MyDrive/Colab Notebooks/brain-tumor-detector/Models/ResUNet-weights_22-12-2025.keras
72/72 ━━━━━━━━━━━━━━━━━━━━ 12s 166ms/step - loss: 0.2861 - tversky: 0.8099 - val_loss: 0.3432 - val_tversky: 0.7583
Epoch 6/60
72/72 ━━━━━━━━━━━━━━━━━━━━ 0s 147ms/step - loss: 0.2626 - tversky: 0.8300
Epoch 6: val_loss improved from 0.34319 to 0.32901, saving model to /content/drive/MyDrive/Colab Notebooks/brain-tumor-detector/Models/ResUNet-weights_22-12-2025.keras
72/72 ━━━━━━━━━━━━━━━━━━━━ 12s 165ms/step - loss: 0.2627 - tversky: 0.8299 - val_loss: 0.3290 - val_tversky: 0.7722
Epoch 7/60
72/72 ━━━━━━━━━━━━━━━━━━━━ 0s 148ms/step - loss: 0.2603 - tversky: 0.8322
Epoch 7: val_loss improved from 0.32901 to 0.26115, saving model to /content/drive/MyDrive/Colab Notebooks/brain-tumor-detector/Models/ResUNet-weights_22-12-2025.keras
72/72 ━━━━━━━━━━━━━━━━━━━━ 12s 166ms/step - loss: 0.2603 - tversky: 0.8322 - val_loss: 0.2612 - val_tversky: 0.8314
Epoch 8/60
72/72 ━━━━━━━━━━━━━━━━━━━━ 0s 144ms/step - loss: 0.2594 - tversky: 0.8330
Epoch 8: val_loss did not improve from 0.26115
72/72 ━━━━━━━━━━━━━━━━━━━━ 11s 157ms/step - loss: 0.2593 - tversky: 0.8331 - val_loss: 0.3317 - val_tversky: 0.7696
Epoch 9/60
72/72 ━━━━━━━━━━━━━━━━━━━━ 0s 144ms/step - loss: 0.2281 - tversky: 0.8595
Epoch 9: val_loss did not improve from 0.26115
72/72 ━━━━━━━━━━━━━━━━━━━━ 11s 157ms/step - loss: 0.2281 - tversky: 0.8595 - val_loss: 0.2777 - val_tversky: 0.8169
Epoch 10/60
72/72 ━━━━━━━━━━━━━━━━━━━━ 0s 150ms/step - loss: 0.2070 - tversky: 0.8770
Epoch 10: val_loss did not improve from 0.26115
72/72 ━━━━━━━━━━━━━━━━━━━━ 12s 163ms/step - loss: 0.2070 - tversky: 0.8770 - val_loss: 0.3404 - val_tversky: 0.7616
Epoch 11/60
72/72 ━━━━━━━━━━━━━━━━━━━━ 0s 142ms/step - loss: 0.2093 - tversky: 0.8744
Epoch 11: val_loss improved from 0.26115 to 0.24138, saving model to /content/drive/MyDrive/Colab Notebooks/brain-tumor-detector/Models/ResUNet-weights_22-12-2025.keras
72/72 ━━━━━━━━━━━━━━━━━━━━ 12s 161ms/step - loss: 0.2092 - tversky: 0.8744 - val_loss: 0.2414 - val_tversky: 0.8488
Epoch 12/60
72/72 ━━━━━━━━━━━━━━━━━━━━ 0s 146ms/step - loss: 0.1751 - tversky: 0.9015
Epoch 12: val_loss improved from 0.24138 to 0.23228, saving model to /content/drive/MyDrive/Colab Notebooks/brain-tumor-detector/Models/ResUNet-weights_22-12-2025.keras
72/72 ━━━━━━━━━━━━━━━━━━━━ 12s 165ms/step - loss: 0.1752 - tversky: 0.9015 - val_loss: 0.2323 - val_tversky: 0.8565
Epoch 13/60
72/72 ━━━━━━━━━━━━━━━━━━━━ 0s 145ms/step - loss: 0.1897 - tversky: 0.8903
Epoch 13: val_loss improved from 0.23228 to 0.20230, saving model to /content/drive/MyDrive/Colab Notebooks/brain-tumor-detector/Models/ResUNet-weights_22-12-2025.keras
72/72 ━━━━━━━━━━━━━━━━━━━━ 12s 164ms/step - loss: 0.1897 - tversky: 0.8903 - val_loss: 0.2023 - val_tversky: 0.8811
Epoch 14/60
72/72 ━━━━━━━━━━━━━━━━━━━━ 0s 148ms/step - loss: 0.1750 - tversky: 0.9015
Epoch 14: val_loss improved from 0.20230 to 0.20201, saving model to /content/drive/MyDrive/Colab Notebooks/brain-tumor-detector/Models/ResUNet-weights_22-12-2025.keras
72/72 ━━━━━━━━━━━━━━━━━━━━ 12s 167ms/step - loss: 0.1750 - tversky: 0.9015 - val_loss: 0.2020 - val_tversky: 0.8810
Epoch 15/60
72/72 ━━━━━━━━━━━━━━━━━━━━ 0s 145ms/step - loss: 0.1699 - tversky: 0.9054
Epoch 15: val_loss improved from 0.20201 to 0.20099, saving model to /content/drive/MyDrive/Colab Notebooks/brain-tumor-detector/Models/ResUNet-weights_22-12-2025.keras
72/72 ━━━━━━━━━━━━━━━━━━━━ 12s 169ms/step - loss: 0.1699 - tversky: 0.9054 - val_loss: 0.2010 - val_tversky: 0.8819
Epoch 16/60
72/72 ━━━━━━━━━━━━━━━━━━━━ 0s 144ms/step - loss: 0.1602 - tversky: 0.9125
Epoch 16: val_loss improved from 0.20099 to 0.19495, saving model to /content/drive/MyDrive/Colab Notebooks/brain-tumor-detector/Models/ResUNet-weights_22-12-2025.keras
72/72 ━━━━━━━━━━━━━━━━━━━━ 12s 162ms/step - loss: 0.1603 - tversky: 0.9125 - val_loss: 0.1949 - val_tversky: 0.8864
Epoch 17/60
72/72 ━━━━━━━━━━━━━━━━━━━━ 0s 146ms/step - loss: 0.1570 - tversky: 0.9150
Epoch 17: val_loss did not improve from 0.19495
72/72 ━━━━━━━━━━━━━━━━━━━━ 12s 160ms/step - loss: 0.1570 - tversky: 0.9150 - val_loss: 0.2004 - val_tversky: 0.8823
Epoch 18/60
72/72 ━━━━━━━━━━━━━━━━━━━━ 0s 144ms/step - loss: 0.1478 - tversky: 0.9213
Epoch 18: val_loss improved from 0.19495 to 0.19305, saving model to /content/drive/MyDrive/Colab Notebooks/brain-tumor-detector/Models/ResUNet-weights_22-12-2025.keras
72/72 ━━━━━━━━━━━━━━━━━━━━ 12s 163ms/step - loss: 0.1478 - tversky: 0.9212 - val_loss: 0.1931 - val_tversky: 0.8875
Epoch 19/60
72/72 ━━━━━━━━━━━━━━━━━━━━ 0s 151ms/step - loss: 0.1435 - tversky: 0.9245
Epoch 19: val_loss improved from 0.19305 to 0.18707, saving model to /content/drive/MyDrive/Colab Notebooks/brain-tumor-detector/Models/ResUNet-weights_22-12-2025.keras
72/72 ━━━━━━━━━━━━━━━━━━━━ 12s 170ms/step - loss: 0.1435 - tversky: 0.9245 - val_loss: 0.1871 - val_tversky: 0.8928
Epoch 20/60
72/72 ━━━━━━━━━━━━━━━━━━━━ 0s 149ms/step - loss: 0.1428 - tversky: 0.9250
Epoch 20: val_loss improved from 0.18707 to 0.18585, saving model to /content/drive/MyDrive/Colab Notebooks/brain-tumor-detector/Models/ResUNet-weights_22-12-2025.keras
72/72 ━━━━━━━━━━━━━━━━━━━━ 12s 167ms/step - loss: 0.1428 - tversky: 0.9250 - val_loss: 0.1859 - val_tversky: 0.8938
Epoch 21/60
72/72 ━━━━━━━━━━━━━━━━━━━━ 0s 149ms/step - loss: 0.1362 - tversky: 0.9296
Epoch 21: val_loss did not improve from 0.18585
72/72 ━━━━━━━━━━━━━━━━━━━━ 12s 162ms/step - loss: 0.1363 - tversky: 0.9296 - val_loss: 0.1887 - val_tversky: 0.8909
Epoch 22/60
72/72 ━━━━━━━━━━━━━━━━━━━━ 0s 145ms/step - loss: 0.1324 - tversky: 0.9322
Epoch 22: val_loss did not improve from 0.18585
72/72 ━━━━━━━━━━━━━━━━━━━━ 11s 158ms/step - loss: 0.1325 - tversky: 0.9321 - val_loss: 0.2548 - val_tversky: 0.8371
Epoch 23/60
72/72 ━━━━━━━━━━━━━━━━━━━━ 0s 143ms/step - loss: 0.1386 - tversky: 0.9278
Epoch 23: val_loss improved from 0.18585 to 0.18294, saving model to /content/drive/MyDrive/Colab Notebooks/brain-tumor-detector/Models/ResUNet-weights_22-12-2025.keras
72/72 ━━━━━━━━━━━━━━━━━━━━ 12s 161ms/step - loss: 0.1386 - tversky: 0.9278 - val_loss: 0.1829 - val_tversky: 0.8961
Epoch 24/60
72/72 ━━━━━━━━━━━━━━━━━━━━ 0s 150ms/step - loss: 0.1323 - tversky: 0.9322
Epoch 24: val_loss did not improve from 0.18294
72/72 ━━━━━━━━━━━━━━━━━━━━ 12s 163ms/step - loss: 0.1323 - tversky: 0.9322 - val_loss: 0.2016 - val_tversky: 0.8809
Epoch 25/60
72/72 ━━━━━━━━━━━━━━━━━━━━ 0s 147ms/step - loss: 0.1302 - tversky: 0.9336
Epoch 25: val_loss improved from 0.18294 to 0.17648, saving model to /content/drive/MyDrive/Colab Notebooks/brain-tumor-detector/Models/ResUNet-weights_22-12-2025.keras
72/72 ━━━━━━━━━━━━━━━━━━━━ 12s 165ms/step - loss: 0.1302 - tversky: 0.9337 - val_loss: 0.1765 - val_tversky: 0.9006
Epoch 26/60
72/72 ━━━━━━━━━━━━━━━━━━━━ 0s 151ms/step - loss: 0.1222 - tversky: 0.9391
Epoch 26: val_loss improved from 0.17648 to 0.16939, saving model to /content/drive/MyDrive/Colab Notebooks/brain-tumor-detector/Models/ResUNet-weights_22-12-2025.keras
72/72 ━━━━━━━━━━━━━━━━━━━━ 12s 170ms/step - loss: 0.1222 - tversky: 0.9391 - val_loss: 0.1694 - val_tversky: 0.9062
Epoch 27/60
72/72 ━━━━━━━━━━━━━━━━━━━━ 0s 147ms/step - loss: 0.1144 - tversky: 0.9442
Epoch 27: val_loss did not improve from 0.16939
72/72 ━━━━━━━━━━━━━━━━━━━━ 12s 160ms/step - loss: 0.1145 - tversky: 0.9442 - val_loss: 0.1742 - val_tversky: 0.9026
Epoch 28/60
72/72 ━━━━━━━━━━━━━━━━━━━━ 0s 143ms/step - loss: 0.1111 - tversky: 0.9464
Epoch 28: val_loss did not improve from 0.16939
72/72 ━━━━━━━━━━━━━━━━━━━━ 11s 156ms/step - loss: 0.1112 - tversky: 0.9464 - val_loss: 0.1856 - val_tversky: 0.8932
Epoch 29/60
72/72 ━━━━━━━━━━━━━━━━━━━━ 0s 141ms/step - loss: 0.1145 - tversky: 0.9442
Epoch 29: val_loss did not improve from 0.16939
72/72 ━━━━━━━━━━━━━━━━━━━━ 11s 153ms/step - loss: 0.1146 - tversky: 0.9442 - val_loss: 0.1869 - val_tversky: 0.8929
Epoch 30/60
72/72 ━━━━━━━━━━━━━━━━━━━━ 0s 144ms/step - loss: 0.1109 - tversky: 0.9464
Epoch 30: val_loss did not improve from 0.16939
72/72 ━━━━━━━━━━━━━━━━━━━━ 11s 156ms/step - loss: 0.1109 - tversky: 0.9464 - val_loss: 0.1718 - val_tversky: 0.9039
Epoch 31/60
72/72 ━━━━━━━━━━━━━━━━━━━━ 0s 144ms/step - loss: 0.1127 - tversky: 0.9453
Epoch 31: val_loss improved from 0.16939 to 0.16876, saving model to /content/drive/MyDrive/Colab Notebooks/brain-tumor-detector/Models/ResUNet-weights_22-12-2025.keras
72/72 ━━━━━━━━━━━━━━━━━━━━ 12s 162ms/step - loss: 0.1127 - tversky: 0.9453 - val_loss: 0.1688 - val_tversky: 0.9060
Epoch 32/60
72/72 ━━━━━━━━━━━━━━━━━━━━ 0s 148ms/step - loss: 0.1066 - tversky: 0.9492
Epoch 32: val_loss did not improve from 0.16876
72/72 ━━━━━━━━━━━━━━━━━━━━ 12s 161ms/step - loss: 0.1066 - tversky: 0.9492 - val_loss: 0.1761 - val_tversky: 0.9008
Epoch 33/60
72/72 ━━━━━━━━━━━━━━━━━━━━ 0s 145ms/step - loss: 0.1026 - tversky: 0.9518
Epoch 33: val_loss did not improve from 0.16876
72/72 ━━━━━━━━━━━━━━━━━━━━ 11s 157ms/step - loss: 0.1027 - tversky: 0.9518 - val_loss: 0.1950 - val_tversky: 0.8855
Epoch 34/60
72/72 ━━━━━━━━━━━━━━━━━━━━ 0s 145ms/step - loss: 0.1071 - tversky: 0.9488
Epoch 34: val_loss did not improve from 0.16876
72/72 ━━━━━━━━━━━━━━━━━━━━ 11s 158ms/step - loss: 0.1072 - tversky: 0.9488 - val_loss: 0.1735 - val_tversky: 0.9030
Epoch 35/60
72/72 ━━━━━━━━━━━━━━━━━━━━ 0s 143ms/step - loss: 0.1063 - tversky: 0.9494
Epoch 35: val_loss did not improve from 0.16876
72/72 ━━━━━━━━━━━━━━━━━━━━ 11s 156ms/step - loss: 0.1063 - tversky: 0.9494 - val_loss: 0.1840 - val_tversky: 0.8947
Epoch 36/60
72/72 ━━━━━━━━━━━━━━━━━━━━ 0s 150ms/step - loss: 0.1029 - tversky: 0.9516
Epoch 36: val_loss improved from 0.16876 to 0.16731, saving model to /content/drive/MyDrive/Colab Notebooks/brain-tumor-detector/Models/ResUNet-weights_22-12-2025.keras
72/72 ━━━━━━━━━━━━━━━━━━━━ 12s 168ms/step - loss: 0.1029 - tversky: 0.9516 - val_loss: 0.1673 - val_tversky: 0.9076
Epoch 37/60
72/72 ━━━━━━━━━━━━━━━━━━━━ 0s 145ms/step - loss: 0.1047 - tversky: 0.9505
Epoch 37: val_loss did not improve from 0.16731
72/72 ━━━━━━━━━━━━━━━━━━━━ 11s 158ms/step - loss: 0.1047 - tversky: 0.9505 - val_loss: 0.1726 - val_tversky: 0.9034
Epoch 38/60
72/72 ━━━━━━━━━━━━━━━━━━━━ 0s 145ms/step - loss: 0.0948 - tversky: 0.9567
Epoch 38: val_loss did not improve from 0.16731
72/72 ━━━━━━━━━━━━━━━━━━━━ 11s 158ms/step - loss: 0.0948 - tversky: 0.9566 - val_loss: 0.1846 - val_tversky: 0.8945
Epoch 39/60
72/72 ━━━━━━━━━━━━━━━━━━━━ 0s 144ms/step - loss: 0.0965 - tversky: 0.9555
Epoch 39: val_loss did not improve from 0.16731
72/72 ━━━━━━━━━━━━━━━━━━━━ 11s 157ms/step - loss: 0.0965 - tversky: 0.9556 - val_loss: 0.1735 - val_tversky: 0.9030
Epoch 40/60
72/72 ━━━━━━━━━━━━━━━━━━━━ 0s 145ms/step - loss: 0.1000 - tversky: 0.9534
Epoch 40: val_loss did not improve from 0.16731
72/72 ━━━━━━━━━━━━━━━━━━━━ 11s 158ms/step - loss: 0.1000 - tversky: 0.9534 - val_loss: 0.1733 - val_tversky: 0.9028
Epoch 41/60
72/72 ━━━━━━━━━━━━━━━━━━━━ 0s 146ms/step - loss: 0.0919 - tversky: 0.9584
Epoch 41: val_loss did not improve from 0.16731
72/72 ━━━━━━━━━━━━━━━━━━━━ 11s 158ms/step - loss: 0.0920 - tversky: 0.9584 - val_loss: 0.1705 - val_tversky: 0.9054
Epoch 42/60
72/72 ━━━━━━━━━━━━━━━━━━━━ 0s 144ms/step - loss: 0.0925 - tversky: 0.9580
Epoch 42: val_loss did not improve from 0.16731
72/72 ━━━━━━━━━━━━━━━━━━━━ 11s 156ms/step - loss: 0.0925 - tversky: 0.9580 - val_loss: 0.1703 - val_tversky: 0.9047
Epoch 43/60
72/72 ━━━━━━━━━━━━━━━━━━━━ 0s 143ms/step - loss: 0.1006 - tversky: 0.9529
Epoch 43: val_loss did not improve from 0.16731
72/72 ━━━━━━━━━━━━━━━━━━━━ 11s 155ms/step - loss: 0.1005 - tversky: 0.9529 - val_loss: 0.1851 - val_tversky: 0.8939
Epoch 44/60
72/72 ━━━━━━━━━━━━━━━━━━━━ 0s 146ms/step - loss: 0.0894 - tversky: 0.9598
Epoch 44: val_loss did not improve from 0.16731
72/72 ━━━━━━━━━━━━━━━━━━━━ 11s 159ms/step - loss: 0.0894 - tversky: 0.9598 - val_loss: 0.1742 - val_tversky: 0.9024
Epoch 45/60
72/72 ━━━━━━━━━━━━━━━━━━━━ 0s 144ms/step - loss: 0.0879 - tversky: 0.9608
Epoch 45: val_loss did not improve from 0.16731
72/72 ━━━━━━━━━━━━━━━━━━━━ 11s 157ms/step - loss: 0.0879 - tversky: 0.9607 - val_loss: 0.1711 - val_tversky: 0.9047
Epoch 46/60
72/72 ━━━━━━━━━━━━━━━━━━━━ 0s 142ms/step - loss: 0.0885 - tversky: 0.9604
Epoch 46: val_loss did not improve from 0.16731
72/72 ━━━━━━━━━━━━━━━━━━━━ 11s 155ms/step - loss: 0.0885 - tversky: 0.9604 - val_loss: 0.1731 - val_tversky: 0.9030
Epoch 47/60
72/72 ━━━━━━━━━━━━━━━━━━━━ 0s 146ms/step - loss: 0.0867 - tversky: 0.9615
Epoch 47: val_loss did not improve from 0.16731
72/72 ━━━━━━━━━━━━━━━━━━━━ 11s 158ms/step - loss: 0.0867 - tversky: 0.9615 - val_loss: 0.1741 - val_tversky: 0.9024
Epoch 48/60
72/72 ━━━━━━━━━━━━━━━━━━━━ 0s 144ms/step - loss: 0.0846 - tversky: 0.9628
Epoch 48: val_loss did not improve from 0.16731
72/72 ━━━━━━━━━━━━━━━━━━━━ 11s 156ms/step - loss: 0.0846 - tversky: 0.9627 - val_loss: 0.1884 - val_tversky: 0.8911
Epoch 49/60
72/72 ━━━━━━━━━━━━━━━━━━━━ 0s 146ms/step - loss: 0.0890 - tversky: 0.9600
Epoch 49: val_loss did not improve from 0.16731
72/72 ━━━━━━━━━━━━━━━━━━━━ 12s 159ms/step - loss: 0.0890 - tversky: 0.9600 - val_loss: 0.1750 - val_tversky: 0.9016
Epoch 50/60
72/72 ━━━━━━━━━━━━━━━━━━━━ 0s 145ms/step - loss: 0.0835 - tversky: 0.9634
Epoch 50: val_loss did not improve from 0.16731
72/72 ━━━━━━━━━━━━━━━━━━━━ 11s 157ms/step - loss: 0.0835 - tversky: 0.9634 - val_loss: 0.1840 - val_tversky: 0.8951
Epoch 51/60
72/72 ━━━━━━━━━━━━━━━━━━━━ 0s 143ms/step - loss: 0.0813 - tversky: 0.9646
Epoch 51: val_loss did not improve from 0.16731
72/72 ━━━━━━━━━━━━━━━━━━━━ 11s 156ms/step - loss: 0.0814 - tversky: 0.9646 - val_loss: 0.1725 - val_tversky: 0.9029
Epoch 52/60
72/72 ━━━━━━━━━━━━━━━━━━━━ 0s 145ms/step - loss: 0.0816 - tversky: 0.9645
Epoch 52: val_loss did not improve from 0.16731
72/72 ━━━━━━━━━━━━━━━━━━━━ 11s 158ms/step - loss: 0.0816 - tversky: 0.9645 - val_loss: 0.1810 - val_tversky: 0.8973
Epoch 53/60
72/72 ━━━━━━━━━━━━━━━━━━━━ 0s 145ms/step - loss: 0.0837 - tversky: 0.9633
Epoch 53: val_loss did not improve from 0.16731
72/72 ━━━━━━━━━━━━━━━━━━━━ 11s 158ms/step - loss: 0.0837 - tversky: 0.9633 - val_loss: 0.1816 - val_tversky: 0.8970
Epoch 54/60
72/72 ━━━━━━━━━━━━━━━━━━━━ 0s 144ms/step - loss: 0.0798 - tversky: 0.9655
Epoch 54: val_loss did not improve from 0.16731
72/72 ━━━━━━━━━━━━━━━━━━━━ 11s 156ms/step - loss: 0.0798 - tversky: 0.9655 - val_loss: 0.1885 - val_tversky: 0.8917
Epoch 55/60
72/72 ━━━━━━━━━━━━━━━━━━━━ 0s 144ms/step - loss: 0.0798 - tversky: 0.9656
Epoch 55: val_loss did not improve from 0.16731
72/72 ━━━━━━━━━━━━━━━━━━━━ 11s 157ms/step - loss: 0.0798 - tversky: 0.9655 - val_loss: 0.1827 - val_tversky: 0.8961
Epoch 56/60
72/72 ━━━━━━━━━━━━━━━━━━━━ 0s 144ms/step - loss: 0.0826 - tversky: 0.9639
Epoch 56: val_loss did not improve from 0.16731
72/72 ━━━━━━━━━━━━━━━━━━━━ 11s 157ms/step - loss: 0.0826 - tversky: 0.9639 - val_loss: 0.1789 - val_tversky: 0.8989
Epoch 56: early stopping

The best model was reached on epoch 36. After that the validation loss did not improve and early stopping rule was reached (patience=20).

if not retrain_model:
    df_his = pd.read_csv(model_base/"training_history_segmodel_22-12-2025.csv", sep=None, engine="python")
    df_his = df_his.apply(pd.to_numeric, errors="coerce")
    history_data = df_his.to_dict(orient="list")
else:
    history_data = history.history  # Keras History object

# ---- plot ----

out_path = image_base/"seg_model_train_history.png"

fig, ax = plt.subplots(figsize=(5, 3))
ax.plot(history_data["loss"], label="train_loss")
ax.plot(history_data["val_loss"], label="val_loss")

ax.set_title("Segmentation Model Loss")
ax.set_ylabel("Loss")

ax.plot(history['accuracy'])
ax.plot(history['val_accuracy'])
ax.set_xlabel("Epoch")
ax.legend(['train_loss','val_loss','train_accuracy','val_accuracy'], loc="center right")

fig.tight_layout()
fig.savefig(out_path, dpi=200, bbox_inches="tight", transparent=True)
plt.close(fig)

display(Image(filename=out_path,width=460))

Assessing the trained segmentation model performance

To assess the performance, the predicted and actual masks of 10 test cases are printed below. The model has not seen this data before.